GitHub Actions fix.

reinforcement-learning/breakout/DQN.py

@@ -275,7 +275,7 @@ def preprocess(self, state):
 
  return state
 
- def observe(self, environment, states, *args): # noqa
+ def observe(self, environment, states, skip=None):
  """
  Observe the environment for n frames.
 
@@ -285,7 +285,7 @@ def observe(self, environment, states, *args): # noqa
  The environment to observe.
  states : torch.Tensor
  The states of the environment from the previous step.
- args
+ skip : int, optional
  To be compatible with the other DQN agents. Added here instead of using ABC.
 
  Returns
@@ -299,6 +299,8 @@ def observe(self, environment, states, *args): # noqa
  done : bool
  Whether the game is terminated.
  """
+ print("Warning: `skip` is not used in `VisionDeepQ.observe`.") if skip is not None else None
+
  action = self.action(states)
 
  done = False

reinforcement-learning/utilities/visualisation/movie.py

@@ -1,29 +1,25 @@
+"""Create a movie of an agent interacting with an environment."""
+
 import cv2
 import torch
 
 
-def create_movie(environment, agent, path, fps=60):
+def create_movie(environment, agent, path="./live-preview.gif", skip=4, fps=50):
  """Created by Mistral Large."""
  initial = agent.preprocess(environment.reset()[0])
  try:
  states = torch.cat([initial] * agent.shape["reshape"][1], dim=1)
  except AttributeError:
  states = initial
 
- try:
- done = False
+ done = False
 
- # Get the dimensions of the first image
- height, width, channels = environment.render().shape
+ height, width, _ = environment.render().shape
+ fourcc = cv2.VideoWriter_fourcc(*"MJPG") # noqa
+ movie = cv2.VideoWriter(path, fourcc, fps, (width, height))
 
- # Create the VideoWriter object
- fourcc = cv2.VideoWriter_fourcc(*"MJPG") # You can change the codec if needed
- video_writer = cv2.VideoWriter(path, fourcc, fps, (width, height))
- while not done:
- _, states, _, done = agent.observe(environment, states)
- video_writer.write(environment.render())
- except Exception as e:
- print(f"Error during image generation or writing: {e}")
- return
+ while not done:
+ _, states, _, done = agent.observe(environment, states, skip)
+ movie.write(environment.render())
 
  cv2.destroyAllWindows()

requirements.txt

@@ -8,8 +8,12 @@ torch
 numpy
 pandas
 scipy
+
+# For visualisation
+# *-----------------------------------------*
 matplotlib
 imageio
+opencv-python
 
 # Library for utilising Apple M chips
 # *-----------------------------------------*